Prove
Citation watch uses the same five prompts every month
By Benoit Maillary · Published 2026-09-14 · Updated 2026-09-15
FixFlow freezes ChatGPT-style prompts on the first GEO persist, then re-runs that same list each month. Cited, not cited, or no longer cited. Change the question and you can no longer read the trend.
If the question changes, the trend is fiction
A consultant who pastes a new ChatGPT prompt every month cannot tell whether the brand lost a citation or whether they simply asked a different thing. Citation watch stores the prompt list on the site. The monthly job re-runs those prompts. Editing the list does not call the model until the next check. The first GEO persist seeds the list only when it is empty. After that, the same five questions are the contract.
Five is a small, honest set. It is not a live share-of-voice grid across every AI product. It is not 200 prompts refreshed daily. You can pin a competitor domain and read You, Competitor, or Neither in the excerpt. That is enough to show a client whether they are still in the answer.
The prompts are ChatGPT-style questions a buyer would type, not keyword strings. “Best SEO audit for a Shopify agency in Lyon” is a prompt. “shopify seo tool” is a keyword. We store the sentence. We re-run the sentence. If you edit prompt 3 in June, July is a new experiment, not a continuation of May. That is why the editor says the list is frozen until the next check.
What the monthly email is for, and what it is not
You get mail when a monthly run shows a lost citation or a newly gained one. The first baseline is not an email. A manual re-check from the dashboard is not an email. Quiet months stay quiet. That rule exists so the inbox is not a scan log.
The GEO Score is how you explain the starting point: five weighted dimensions, 0-100. Citation watch is how you explain the month: the same prompts, cited or not cited anymore. Pro exists for the second number. A one-shot GEO score is not a retainer. See plans.
A lost citation is not a guarantee that traffic dropped. It is a guarantee that this prompt, on this run, no longer named you. That is still a fact you can put in a deck. Inventing a percentage of “AI market share” from five prompts would be a lie. We will not add that number.
You, competitor, or neither
The excerpt parser looks at hosts, not vibes. If the answer names your domain, the row is You. If it names the pinned competitor and not you, the row is Competitor. If it names nobody you care about, the row is Neither. Pinning is optional. Without a pin, you still get cited / not cited. With a pin, you can show a client who replaced them.
That is still not a ranking. Two brands can appear in one answer. We record what the excerpt gave us on that run. The next month uses the same prompts so you can see a flip. If you want a different competitor, change the pin. Do not change the five questions unless you accept a break in the series.
Where this sits in find, repair, prove
FixFlow’s job is to be found, repair, and prove. The ordered technical list is the find-and-repair order. The paste pack is how you ship without handing over the CMS login. Citation watch is the prove step for AI answers. The features page is the roster of names. This page is the rule for the prompts: freeze them, re-run them, read cited or not.
If you want a free score on a CMS or a sector URL first, start from the audit hub. The blog is the method. The scan is the number on your URL. The monthly email is only the delta. The frozen list is the contract you can show in a QBR without inventing a new question each time.
What a first month looks like
Week one: you persist GEO, the five prompts are seeded if the list was empty, and you have a baseline. No email. Week four-ish, when the calendar month changes, the same five run again. If nothing flipped, silence. If you dropped off a prompt, mail. If you appeared on a prompt you missed last month, mail. That is the whole product surface for citation watch.
Consultants who want a weekly ChatGPT screenshot can still take one. We will not store it as a data point. Weekly screenshots of different questions are how retainers become folklore. Monthly runs of the same questions are how they become a line in a report. If the client insists on a weekly ritual, run the same five prompts yourself. Do not add a sixth and call it the same series.